1 Introduction

Genome wide association studies (GWAS) can reveal important genotype to phenotype associations, however, data quality and interpretability issues must be addressed. The GWAX approach enables rational ranking, filtering and interpretation of GWAS via metrics, methods, and interactive visualization. Each inferred gene-to-trait association is evaluated for confidence and relevance, with scores solely derived from aggregated statistics, linking a protein-coding gene and phenotype. Applicability and thresholds will depend on use cases.

Updated to GWAS Catalog v1.0.2, Sept 30, 2018.
Workflow designed for integration with TCRD & Pharos
iCite Relative Citation Ratio (RCR) integration (in development)
EFO subclass linking of related studies (in development)
Prototype RShiny app: http://unmtid-shinyapps.net/gwax/

Issues/to-do:

Confidence scoring for SNP-gene mappings (“auto-semi-fine mapping” possible?)
Confidence higher for reported or mapped? Current guess: reported.
BETA not comparable for different units, thus currently only OR used for GWAX.
Polygenic diseases/traits: Filter if less suitable for GWAS-based drug target discovery use cases.

1.1 About NHGRI-EBI GWAS Catalog

GWAS Catalog (http://www.ebi.ac.uk/gwas/) studies each have a study_accession. Also are associated with a publication (PubMedID), but not uniquely. See https://www.ebi.ac.uk/gwas/docs/fileheaders.

Some key definitions:

`REPORTED GENE(S)`*: Gene(s) reported by author

`MAPPED GENE(S)`: Gene(s) mapped to the strongest SNP. If the SNP is located
within a gene, that gene is listed. If the SNP is intergenic, the upstream
and downstream genes are listed, separated by a hyphen.

`OR or BETA`: Reported odds ratio or beta-coefficient associated with
strongest SNP risk allele. Note that if an OR <1 is reported this is
inverted, along with the reported allele, so that all ORs included in
the Catalog are >1. Appropriate unit and increase/decrease are included
for beta coefficients.

Welter D, MacArthur J, Morales J, Burdett T, Hall P, Junkins H, Klemm A, Flicek P, Manolio T, Hindorff L, and Parkinson H. The NHGRI GWAS Catalog, a curated resource of SNP-trait associations. Nucleic Acids Research, 2014, Vol. 42 (Database issue): D1001-D1006.

## [1] "Wed Jun  5 10:12:59 2019"

2 Read files

From GWAS Catalog, TCRD, and EFO.

3 Genome Wide Association Studies (GWAS)

3.0.1 Counts by year

## [1] "Studies total: 5774 ; accessions: 5774 ; traits: 3472 ; PMIDs: 3596"

### Laboratory platforms

Grouped by vendor (first in list if multiple), though technologies may have evolved for a given vendor.

3.0.2 Journals

Top journals by N_gwas
JOURNAL	N_gwas	N_pmid	N_assn
Nat Genet	797	522	16177
PLoS One	383	226	4966
PLoS Genet	352	178	6419
Hum Mol Genet	352	243	4154
Nat Commun	307	128	7928
Am J Hum Genet	162	85	3238
Sci Rep	149	79	1003
Mol Psychiatry	138	98	1970
Circ Cardiovasc Genet	88	54	623
Am J Med Genet B Neuropsychiatr Genet	80	47	1070
Diabetes	72	39	463
Hum Genet	72	51	1693

4 iCite Relative Citation Ratio (RCR)

iCite annotations from iCite API, with all PMIDs from GWASCatalog. New publications may lack iCite RCR. Should we impute RCR=median as reasonable prior?

## [1] "N_pmid = 5774"                          
## [2] "mean = 4.2 ; median = 2.0 ; max = 161.8"
## [3] "90%ile = 8.5"                           
## [4] "(Plot truncated at 25.)"

5 Associations (SNP to trait)

## [1] "Associations total: 87601 ; SNPs: 59580 ; traits: 2970 ; PMIDs: 3110"

6 SNP to gene mappings

GSYMB, MAPPED_GENE fields may include chromosomal locations or be “intergenic”.
Chromosomal locations (LOC*) may be non-coding RNA, etc.
Mappings within gene stronger associations than upstream or downstream.

## [1] "snp2gene: total associations: 266000 ; studies: 4902 ; snps: 60099 ; genes: 22392 ; intergenic associations: 5145 ; chromosomal location associations: 38630"

REPORTED_OR_MAPPED	N
reported	116747
mapped_upstream	38759
mapped_downstream	38759
mapped_within	71735

7 Gene counts

## [1] "Studies: 4902"

## [1] "MAPPED_GENE values: 22750"

## [1] "REPORTED_GENE values: 18396"

## [1] "TCRD targets: 19947 ; geneSymbols: 19736"

## [1] "GSYMBs mapped to TCRD: 13725"

## [1] "Tbio: 7995"  "Tchem: 1513" "Tclin: 512"  "Tdark: 3831"

8 Gene-SNP-Study-Trait (G2T) associations

g2t should have one row for each gene-snp-study-trait association.

## [1] "GTs with pvalue_mlog, g2t: 242998 ; genes: 20681 ; traits: 1601"

## [1] "GTs with or_or_beta, g2t: 242998 ; genes: 20681 ; traits: 1601"

## [1] "GTs with oddsratio, g2t: 60065 ; genes: 12177 ; traits: 913"

## [1] "GTs with beta, g2t: 137631 ; genes: 12407 ; traits: 767"

9 Traits and EFO

EFO = Experimental Factor Ontology. Includes Orphanet, PO, Mondo and Uberon classes. TSV from source OWL.

## [1] "EFO total classes: 29085"

EFO sources, top 5
Ontology	N_in_gwas	N_total
EFO	1719	9642
GO	67	348
HP	46	474
Orphanet	40	5989
CHEBI	2	1317

9.1 GWAS trait-subclass relationships

## [1] "EFO classes: 29085 ; total subclass relationships: 48420"

## [1] "GWAS trait-subclass pairs: 1280"

Top EFO trait-subclass pairs
trait_id	trait_name	subclass_id	subclass_name	trait_N_gwas	subclass_N_gwas
EFO_0004340	body mass index	EFO_0005937	longitudinal BMI measurement	96	12
EFO_0004340	body mass index	EFO_0005935	overweight body mass index status	96	3
EFO_0004340	body mass index	EFO_0005936	underweight body mass index status	96	2
EFO_0004340	body mass index	EFO_0005851	height-adjusted body mass index	96	1
EFO_0004340	body mass index	EFO_0007041	obese body mass index status	96	1
EFO_0000692	schizophrenia	EFO_0004609	treatment refractory schizophrenia	94	6
EFO_0000305	breast carcinoma	EFO_1000649	estrogen-receptor positive breast cancer	77	14
EFO_0000305	breast carcinoma	EFO_1000650	estrogen-receptor negative breast cancer	77	10
EFO_0000305	breast carcinoma	EFO_1002010	TP53 Positive Breast Carcinoma	77	1
EFO_0000249	Alzheimer’s disease	EFO_1001870	late-onset Alzheimers disease	74	2
EFO_0004612	high density lipoprotein cholesterol measurement	EFO_0007805	HDL cholesterol change measurement	70	3
EFO_0000270	asthma	EFO_0004591	childhood onset asthma	60	12
EFO_0000270	asthma	EFO_1002011	adult onset asthma	60	3
EFO_0004530	triglyceride measurement	EFO_0007681	triglyceride change measurement	59	5
EFO_0005842	colorectal cancer	EFO_1000657	rectum cancer	59	2
EFO_0005842	colorectal cancer	EFO_1001480	metastatic colorectal cancer	59	2
EFO_0001663	prostate carcinoma	EFO_0000196	metastatic prostate cancer	56	2
EFO_0004611	low density lipoprotein cholesterol measurement	EFO_0007804	LDL cholesterol change measurement	55	4
EFO_0000685	rheumatoid arthritis	EFO_0003898	ankylosing spondylitis	46	7
EFO_0000685	rheumatoid arthritis	EFO_0003778	psoriatic arthritis	46	4
EFO_0003923	bone density	EFO_0007701	spine bone mineral density	45	15
EFO_0001645	coronary heart disease	EFO_0000378	coronary artery disease	45	14
EFO_0003923	bone density	EFO_0007702	hip bone mineral density	45	9
EFO_0003923	bone density	EFO_0007933	radius bone mineral density	45	3

9.2 GWAS studies related by EFO subclass links.

## [1] "GWAS studies related by EFO-subclass: 1289"

10 GENE-TRAIT stats

Read gt_stats.tsv, built by gwax_gt_stats.R for GWAX. Statistics designed to weigh evidence aggregated across studies, for each gene-trait association.

TO DO: Add EFO subclass based evidence aggregation scores.

## [1] "nrow(gt) = 33977"

10.0.1 Focus on traits with more data and evidence, likely high scientific interest.

Most-highly studied traits
trait	trait_ids	N_genes
schizophrenia	EFO_0000692	1716
intelligence	EFO_0004337	1232
autism spectrum disorder	EFO_0003756	886
prostate carcinoma	EFO_0001663	549
inflammatory bowel disease	EFO_0003767	527
systemic lupus erythematosus	EFO_0002690	507
age at onset	EFO_0004847	481
Crohn’s disease	EFO_0000384	468
lung carcinoma	EFO_0001071	451
Alzheimer’s disease	EFO_0000249	433
bipolar disorder	EFO_0000289	422
asthma	EFO_0000270	418

11 Prototype GWAX web app.

11.1 Plot single trait with all associated genes.

X-axis: Evidence (n_study)
Y-axis: Effect (median(OR))
Other considerations:
n_traits_this_gene - Normally prefer low value, but depends on other traits, semantics/ontology.
n_snp - How many SNPs? But is more or fewer better?
pval_median - Interpretation may be a challenge.
Note/issue: Ignoring beta values because not comparable, due to varying units. (But maybe can be aggregated if units agree.)

Color unmapped gray.

Plot for a selected trait:

## [1] "http://www.ebi.ac.uk/efo/EFO_0002508: Parkinson's disease"

11.1.1 Pareto filter

Selects N non-dominated solutions on 2D multi-objective boundary.

11.1.2 Filter for (1) Evidence (n_study) and (2) effect (or_median).

11.1.3 Plot

Top hits:

GWAX: Parkinson’s disease, top hits
gsymb	name	fam	tdl	n_study	rcras	or_median	pvalue_mlog_median
C16orf75	NA	NA	NA	7	24.304	1.270	43.000
MHC	NA	NA	NA	6	17.928	1.240	16.398
GAK	Cyclin-G-associated kinase	Kinase	Tchem	5	19.635	1.260	50.000
KIAA1267	NA	NA	NA	3	9.586	1.272	28.000
CCHCR1	Coiled-coil alpha-helical rod protein 1	NA	Tbio	3	9.508	1.390	6.000
PRDM15	PR domain zinc finger protein 15	TF; Epigenetic	Tdark	3	12.415	1.143	23.097
ITGA8	Integrin alpha-8	NA	Tbio	3	10.556	1.330	5.523
GLTSCR1L	GLTSCR1-like protein	NA	Tdark	3	10.059	1.190	7.523
HLA-DRA	HLA class II histocompatibility antigen, DR alpha chain	NA	Tbio	2	5.25	1.310	9.301
PLEK	Pleckstrin	NA	Tbio	2	4.247	1.350	5.699

GWAS Explorer (GWAX)

Jeremy Yang